Shawn Wang: prediction

Transcript context

…Look at the human society, right? We specialize. And you feel really good about someone specializing doing something really well, right? And that's how our way evolved from ancient times. We're all journalists. We do everything. Now we heavily specialize in different domains. So my prediction is in the AI model space, it will happen also. Except for the bitter lesson. You get short-term gains by having specialists, domain specialists, and then someone just needs to train like a 10x bigger model on 10x more inference, 10x more data, 10x more model perhaps, whatever the current scaling law is. And then it supersedes all the individual models because of some generalized intelligence slash world knowledge. I think that is the core insight of the GPTs, the GPT-123 networks. Right. But the training scaling law is because you have an increasing amount of data to train from. And you can do a lot of compute. So I think on the data side, we're approaching the limit. And the only data to increase that is synthetic generated data. And then there's like what is the secret sauce there, right? Because if you have a very good large model, you can generate very good synthetic data and then continue to improve quality. So that's why I think in OpenAI, they are shifting from the training scaling law into…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.